Semantic Segmentation on BDD100K night
50.42mIoUNightLab-HDM
Evaluation results
| Method | Links | |
|---|---|---|
| NightLab-HDMBackbone=Swin-Base, Resolution=720x1280, Training Data=With BDD100K daytime data augmentation2022.04 | 50.42 | |
| Night Lab-HDMAdaptation Approach=Dual-level segmentation, Network=UPerNet-Swin-DeformConv, Training Protocol=Joint train with BDD100K-Day2022.04 | 50.24 | |
| Night Lab-RDNAdaptation Approach=Dual-level segmentation, Network=UPerNet-Swin-DeformConv, Training Protocol=Joint train with BDD100K-Day2022.04 | 49.81 | |
| NightLab-RDNBackbone=Swin-Base, Resolution=720x1280, Training Data=With BDD100K daytime data augmentation2022.04 | 49.81 | |
| UPerNetAdaptation Approach=Segmentation, Network=UPerNet-Swin, Training Protocol=Joint train with BDD100K-Day2022.04 | 48.52 | |
| Night Lab-BAdaptation Approach=Segmentation, Network=UPerNet-Swin-DeformConv, Training Protocol=Joint train with BDD100K-Day2022.04 | 48.52 | |
| NightLab-BaselineBackbone=Swin-Base, Resolution=720x1280, Training Data=With BDD100K daytime data augmentation2022.04 | 48.52 | |
| SingleHDRAdaptation Approach=Image Enhancement, Network=UPerNet-Swin, Training Protocol=Joint train with BDD100K-Day2022.04 | 48.32 | |
| AdaptSegAdaptation Approach=Network Adaptation, Network=UPerNet-Swin, Training Protocol=Joint train with BDD100K-Day2022.04 | 48.32 | |
| DANNetAdaptation Approach=Network Adaptation, Network=UPerNet-Swin, Training Protocol=Joint train with BDD100K-Day2022.04 | 48.25 | |
| UPer-SwinBackbone=Swin-Base, Resolution=720x1280, Training Data=With BDD100K daytime data augmentation2022.04 | 48.04 | |
| UPer-ViTBackbone=ViT, Resolution=720x1280, Training Data=With BDD100K daytime data augmentation2022.04 | 47.81 | |
| UPerNetBackbone=Res101, Resolution=720x1280, Training Data=With BDD100K daytime data augmentation2022.04 | 47.68 | |
| PSPNetBackbone=Res101, Resolution=720x1280, Training Data=With BDD100K daytime data augmentation2022.04 | 46.24 | |
| NightLab (DeeplabV3+)Backbone=Res101, Resolution=720x1280, Training Data=With BDD100K daytime data augmentation2022.04 | 45.11 | |
| HRNetV2Backbone=HRNet-W48, Resolution=720x1280, Training Data=With BDD100K daytime data augmentation2022.04 | 44.32 | |
| DeeplabV3+Backbone=Res101, Resolution=720x1280, Training Data=With BDD100K daytime data augmentation2022.04 | 43.44 | |
| DANetBackbone=Res101, Resolution=720x1280, Training Data=With BDD100K daytime data augmentation2022.04 | 42.64 | |
| CoDABackbone=SegFormer, Scene Specificity=Scene-Agnostic2024.03 | 41.9 | |
| MICBackbone=SegFormer, Scene Specificity=Scene-Agnostic2024.03 | 41.3 | |
| Refign-HRDASource Domain=Cityscapes, Input Protocol=Non-standard resizing protocol2022.07 | 40.6 | |
| SePiCoBackbone=SegFormer, Scene Specificity=Scene-Agnostic2024.03 | 40.6 | |
| CycleGANAdaptation Approach=Image Translation, Network=UPerNet-Swin, Training Protocol=Joint train with BDD100K-Day2022.04 | 39.64 | |
| HRDASource Domain=Cityscapes, Input Protocol=Non-standard resizing protocol2022.07 | 39.1 | |
| Pix2PixHDAdaptation Approach=Image Translation, Network=UPerNet-Swin, Training Protocol=Joint train with BDD100K-Day2022.04 | 38.67 | |
| SePiCo (DAFormer)Source Domain=Cityscapes, Input Protocol=Non-standard resizing protocol2022.07 | 36.9 | |
| Night Lab-HDMAdaptation Approach=Dual-level segmentation, Network=UPerNet-Swin-DeformConv, Training Protocol=Single train2022.04 | 35.41 | |
| NightLab-HDMBackbone=Swin-Base, Resolution=720x1280, Training Data=Night-only2022.04 | 35.41 | |
| MGCDABackbone=ResNet-1012020.05 | 34.9 | |
| MGCDABaseline Architecture=RefineNet2022.05 | 34.9 | |
| MCGDABackbone=RefineNet, Scene Specificity=Scene-Specialized2024.03 | 34.9 | |
| Night Lab-RDNAdaptation Approach=Dual-level segmentation, Network=UPerNet-Swin-DeformConv, Training Protocol=Single train2022.04 | 34.13 | |
| NightLab-RDNBackbone=Swin-Base, Resolution=720x1280, Training Data=Night-only2022.04 | 34.13 | |
| DAFormerBackbone=SegFormer, Scene Specificity=Scene-Agnostic2024.03 | 33.9 | |
| GCMABackbone=ResNet-1012020.05 | 33.2 | |
| GCMABaseline Architecture=RefineNet2022.05 | 33.2 | |
| GCMABackbone=RefineNet, Scene Specificity=Scene-Specialized2024.03 | 33.2 | |
| CCDistillBaseline Architecture=RefineNet2022.05 | 33 | |
| Night Lab-BAdaptation Approach=Segmentation, Network=UPerNet-Swin-DeformConv, Training Protocol=Single train2022.04 | 32.37 | |
| NightLab-BaselineBackbone=Swin-Base, Resolution=720x1280, Training Data=Night-only2022.04 | 32.37 | |
| UPerNetAdaptation Approach=Segmentation, Network=UPerNet-Swin, Training Protocol=Single train2022.04 | 31.74 | |
| UPer-SwinBackbone=Swin-Base, Resolution=720x1280, Training Data=Night-only2022.04 | 31.74 | |
| SingleHDRAdaptation Approach=Image Enhancement, Network=UPerNet-Swin, Training Protocol=Single train2022.04 | 31.64 | |
| NightLab (DeeplabV3+)Backbone=Res101, Resolution=720x1280, Training Data=Night-only2022.04 | 31.27 | |
| UPerNetBackbone=Res101, Resolution=720x1280, Training Data=Night-only2022.04 | 30.88 | |
| UPer-ViTBackbone=ViT, Resolution=720x1280, Training Data=Night-only2022.04 | 30.74 | |
| DANNet(RefineNet)Baseline Architecture=RefineNet2022.05 | 30.3 | |
| DeeplabV3+Backbone=Res101, Resolution=720x1280, Training Data=Night-only2022.04 | 30.11 | |
| PSPNetBackbone=Res101, Resolution=720x1280, Training Data=Night-only2022.04 | 29.96 | |
| HRNetV2Backbone=HRNet-W48, Resolution=720x1280, Training Data=Night-only2022.04 | 29.86 | |
| DANetBackbone=Res101, Resolution=720x1280, Training Data=Night-only2022.04 | 29.46 | |
| DMAdaBackbone=ResNet-1012020.05 | 28.3 | |
| DMAdaBaseline Architecture=RefineNet2022.05 | 28.3 | |
| RefineNetBackbone=ResNet-101, Adaptation Setting=Daytime-trained baseline2020.05 | 26.6 | |
| DeepLab-v2Backbone=ResNet-101, Adaptation Setting=Daytime-trained baseline2020.05 | 22.9 | |
| BDL-Cityscapes→DZ-nightBackbone=ResNet-101, Adaptation Setting=Cityscapes to Dark Zurich-night adaptation2020.05 | 22.8 | |
| BDL-Cityscapes→DZ-nightBaseline Architecture=DeepLab-v22022.05 | 22.8 | |
| ADVENT-Cityscapes→DZ-nightBackbone=ResNet-101, Adaptation Setting=Cityscapes to Dark Zurich-night adaptation2020.05 | 22.6 | |
| ADVENT-Cityscapes→DZ-nightBaseline Architecture=DeepLab-v22022.05 | 22.6 | |
| AdaptSegNet-Cityscapes→DZ-nightBackbone=ResNet-101, Adaptation Setting=Cityscapes to Dark Zurich-night adaptation2020.05 | 22 | |
| AdaptSegNet-Cityscapes→DZ-nightBaseline Architecture=DeepLab-v22022.05 | 22 | |
| AdaptSegBackbone=DeepLab-v2, Scene Specificity=Scene-Agnostic2024.03 | 22 | |
| RefineNet-CityscapesBaseline Architecture=RefineNet, Backbone=ResNet-1012022.05 | 20.4 | |
| UDAclustering-Cityscapes→DZ-nightBaseline Architecture=DeepLab-v22022.05 | 20 | |
| DeepLab-v2-CityscapesBaseline Architecture=DeepLab-v2, Backbone=ResNet-1012022.05 | 17.3 |